Mapping and understanding humins coking during carbohydrate conversion is crucial for improving solid catalyst systems. However, methods for mapping coking are still in need of development. In this study, a Lattice Boltzmann method (LBM) based Back-propagation artificial neural network (BPANN) reduced-order model (ROM) is developed to map humins distribution during the conversion process. The ROM reveals 3 configurations of intra-particle coking distribution (surface-focus, middle-layer-focus and central focus coking). Results indicate that the surface-focus configuration leads to a decreasing trend in the macroscopic coke accumulation rate with reaction cycles,, especially under extreme conditions (surface humins/central humins > 10). Depending on the proportion of HMF-derived humins and the Thiele modulus of the substrate, the proportion of three configurations varies with catalyst load, pellet size, and substrate concentration. The extreme surface-focus coking configuration can be determined by calculating the ideal humins density on pellet surface. Corresponding catalytic system design strategies are proposed.